12 Bold Assertions on How AI Is Redefining Software

Thorsten Ball, core developer of the Zed editor and recent Amp Labs member, posted an X article titled "Software After Software" that presents twelve numbered assertions about the software industry in the AI era.

1. Everything Is Changing

We are in the middle of a massive transformation; almost every assumption about software and business operations must be re‑examined. The shift will take years, and the end state is unclear, but staying proactive is preferable to watching.

2. Model Progress Is Still Underestimated

Even if you think you are accounting for exponential improvement, you are likely still under‑estimating it. Models do not need to be perfect to disrupt old software economics; they only need to be better than average.

Why add a stabilizer to someone who never wobbles? Why give a perfect typist a spell‑checker?

The implication is that many current practices aimed at helping models will become unnecessary. Output quality will depend entirely on input quality, not model capability; code review demand will fade, and expression‑level errors will disappear.

Every eight weeks you should give the model more autonomy, otherwise you’ll get stuck at a low point on the curve.

3. Most Future Code Will Be Written by Agents

Side‑bars for assistants are dead; agents are no longer auxiliary. They will run when no one is at the computer, for longer periods and with far less supervision—beyond most people’s imagination.

An agent forced to work like a human is a wasted agent.

The work unit shifts from "code to write" to "task to delegate".

4. The Bottleneck Has Shifted

Generating syntactically correct code is now trivial. The remaining challenges are engineering‑level: prioritization, ordering, and trade‑offs.

Review will move from code to decisions.

5. Forty Years of Software Development Practices Are Dead

Software as a profession has been built on the assumption that "writing code is hard and error‑prone." As an industry, it has relied on the belief that "code is scarce." Both assumptions no longer hold.

6. Old Processes No Longer Apply

Existing rituals and processes—planning, prioritization, hand‑offs, reviews—were created for a world where development was slow, expensive, error‑prone, and tied to human effort.

Why spend an hour prioritizing when the task can be done in thirty minutes?

Why consider prototyping when you can simply describe the solution?

Why wait for a person to review when you can launch five agents to review in parallel?

Why test a single idea when you can try them all in parallel and discard the four that fail without hurting anyone’s ego?

AI is not just an accelerator for X; it changes whether X should exist at all.

7. Software Changes Its Form

Software has so far been built for human use. In the future, most software will be built for agents.

More software will be generated just‑in‑time rather than built ahead of time.

The line between "using software" and "building software" will blur, possibly disappearing.

8. Value Shifts From Code

When an agent can perform X, the software that "does X" loses value. The act of "someone else writing the code for you" becomes less valuable, and coding‑centric workflows become less lucrative.

What becomes more valuable are data, permissions, distribution channels, trust, compliance, regulatory standing, and physical assets .

Suppliers whose moat relies on "customers cannot build it themselves" will be squeezed.

Eventually, software companies themselves will have to admit this, as their business models were built on the old scarcity.

9. Winners Re‑organize Around the Model

Embedding a model into existing systems, org charts, and processes is insufficient. Forcing agents to work like humans wastes them.

The winners of this transformation are those who reorganize around the model.

A small team with strong judgment plus many agents will outperform a large team trying to bolt AI onto old processes.

Capability to adapt matters more than headcount.

10. Every Serious Organization Needs a Frontier Squad

Large companies cannot ignore this transformation; they cannot learn the future through committees.

They need a small autonomous team stationed at the frontier, building around the model:

Its purpose is not to add AI to old ways of working.

Its purpose is to discover new ways of working and pull the company toward them.

It must be close enough to real systems, real data, real constraints, and real consequences.

Its output is not just software, but also people and practices.

11. Engineer Roles Change

Software engineers will not disappear, but their role will evolve.

Software engineering will shift from code to product thinking, system shaping, trade‑off judgment, and business outcomes.

The most valuable engineers will be system thinkers who can drive commercial value—this was true in the past and will remain true.

12. Therefore

We build for an intelligent, abundant, cheap world.

We optimize learning what that world looks like.

We do not retain old software processes out of habit.

We do not wait for the end state to become obvious; the best strategy is to play the game now.

We go to the frontier first, because that is where we learn about the future.

This is the creed of Thorsten Ball and Amp Labs.

Whether you fully agree or not, these twelve points provide a clear framework for thinking about how the value chain, organization, and engineer roles will be redistributed when code is no longer scarce.

For teams still using old processes to manage AI tools, statements 6 and 9 are worth rereading.

Original post: https://x.com/thorstenball/status/2059304318055150066

Reference link: Amp Labs – https://amplabs.com/